Enhancing Resilience in Human-Reservoir Systems with NLP and AI Frameworks

The resilience of a large-scale water infrastructure system to cascading effects is
fundamentally dependent on the interdependencies of its components within the
infrastructure network. These interdependencies—which means that the states of
two or more infrastructure components are tightly interrelated through mechanisms
such as physical connection, geographical proximity, and information relay—can
cause a localized event to spread into a system-wide event. Of these, logical
interdependencies remain poorly understood. Little is known about how two
infrastructures affect the state of each other through human decisions and how such
logical connections can be detected and measured. In this study, we tackled this
gap by conducting an applied case study on the Lake Mendocino Reservoir in
California, USA. Crucially, our approach focuses on reservoir institutions (rules)
that structure human decisions around reservoir systems. Reservoir management
relies heavily on operational rules and regulations, but climate change demands
more adaptive and discretionary decision-making by operators. This may further
introduce logical interdependencies in a reservoir system. We develop a novel
framework that integrates Institutional Analysis using Large Language Models to
advance Natural Language Processing (NLP) techniques and Bayesian Network
Modeling to systematically analyze and quantify risk associated with logical
interdependencies. We aim to improve decision-making and risk management in
reservoir operations. This research provides essential insights into enhancing the
resilience of water management infrastructures, particularly in the face of climate
change.